New batches of cars Tesla Model S uses a supercomputer Nvidia Drive PX 2, but apparently a non-standard configuration
Tesla cars made a kind of revolution. Or rather, several. First, they changed the majority opinion about electric cars. Second, it provided virtually full autopilot. Third, Tesla cars use the principles of the IT world, when through regular updates ON the car can get improved or even completely new function.
Another interesting approach Tesla to configure their car. The fact that the components responsible for processing data from sensors of the vehicle (including the work of the autopilot), for years of existence on the market models of the company changed several times already. And Tesla is not particularly spread about it.
As it turned out, new parties cars Tesla uses automotive supercomputers Nvidia Drive PX 2.
One of the owners of a brand new Tesla Model S decided to dismantle part of the car to get to this computer. Moreover, it was dismantled and the PC that allowed us to learn about the presence of Nvidia Drive PX 2 GPU GP106. We will remind, it is based on the GeForce GTX 1060! Here you can see a full-fledged graphics processor, but also to note the different design of the PCB. Also, judging by the photos the back side, under the second heatsink, no GPU, no. Apparently, there just and is located a single-chip system Tegra Parker.
One version of the Nvidia Drive PX 2 just involves the use of a single SoC Tegra X2 and a single GPU Pascal, except with a different PCB design. This suggests that especially for Tesla created such version of the Nvidia Drive PX 2, which you can see in the photo. It can also be noted that this modification of supercomputer Nvidia positioned itself as the basis for the advanced cruise control, but not for driving.
In addition, the specifications of the GPU in this version involve the presence of only 256 CUDA cores. That is, either at Tesla special version with a much more efficient GPU or GP106 in this case severely curtailed, although it is unclear why bother to use such a productive solution, if Nvidia has significantly more budget.